| In modern industry,many productive processes exhibit complex nonlinearities and varying dynamic characteristics.Linear parameter-varying systems have drawn a great deal of attention because of their simple linear structure and the ability to describe nonlinear or time-varying properties.The parameters of the linear parameter-varying system are time-varying nonlinear functions of the scheduling variables,but the dynamic relationship between the input and the output is linear.To the best of the author’s knowledge,most of the existing research results assume that the distribution of the noise is known in advance,which is inaccessible in practice since the noise is unmeasurable.To this end,this dissertation develops robust recursive identification methods,which don’t require the priori distribution of the noise and can achieve real-time estimation of the parameters based on the correntropy theory.The main work of this dissertation is as follows.For the linear parameter-varying controlled autoregressive system with the non-Gaussian noise,based on the maximum correntropy criterion,a robust multiinnovation gradient identification method is derived by using the robust performance of the correntropy against noise disturbance,which reduces the influence of the nonGaussian noise on parameter estimation.To deal with the identification problem of the linear parameter-varying system under the multi-peak distribution noise,a robust multi-innovation gradient identification method is proposed by combining the multiple correntropy with different kernel bandwidths based on the maximum multi-kernel correntropy criterion.For the linear parameter-varying output error system with the non-Gaussian noise,based on the maximum correntropy criterion,a robust recursive least squares identification method is proposed.To improve the computational efficiency,a robust recursive least squares method is presented by decomposing the identification model and using the hierarchical identification principle.To increase the flexibility of and to improve the accuracy of the parameter estimation,a robust recursive least squares identification method is derived based on the maximum multi-kernel correntropy.For the linear parameter-varying output error system with the non-Gaussian noise,based on the maximum correntropy criterion,a robust Newton recursive algorithm is derived by using the Newton search principle.Moreover,based on the maximum asymmetric correntropy criterion,a robust Newton recursive algorithm is proposed by introducing the asymmetric correntropy,which reduces the sensitivity of the asymmetric distribution noise to parameter estimation. |
